Wetlands are artificial ponds, designed to filter and purify running water through the contact with plant stems and roots. Wetland layouts are traditionally designed by experts through a laborious and time-consuming procedure: in principle, small patches of vegetation with purifying properties are tentatively placed, then the resulting water flow is verified by fluid dynamics simulators and when a satisfying outcome is reached, the wetland final layout is decided. This paper proposes to automate wetland design exploiting an evolutionary algorithm: a population of candidate solutions is cultivated by the evolutionary core, and their efficiency is evaluated using a state-of-the-art fluid-dynamics simulation framework. Experimental results show that the results obtained by the proposed approach are qualitatively comparable with those provided by experts, despite the complete absence of human intervention during the optimization process.